Claude + Grafana: What the Integration Can (and Can't) Do in 2026
Yes — Grafana ships an official MCP server you can connect to Claude. Just be precise about what it is: Grafana is not in Anthropic’s connectors directory (claude.com/connectors/grafana returns a 404). What exists is an official vendor MCP server in two flavors — the mature self-hosted grafana/mcp-grafana and a brand-new hosted Grafana Cloud MCP server at https://mcp.grafana.com/mcp (public preview since GrafanaCON 2026). Wire either up and Claude can query dashboards, run datasource queries, and read alerts from a chat. The catch is the usual one: it only works inside a conversation you start. There are no triggers, nothing watches your metrics, and nothing runs while you’re away.
Here’s exactly what the integration does, how to turn it on, where the limits bite, and what to use if you want Grafana work that runs on its own.
What the Grafana MCP server does
The server connects Claude to your Grafana instance and exposes 50+ tools across observability, so you drive investigation in natural language.
In practice, Claude can:
- Search and read dashboards — find a dashboard, pull its panels, render a panel image.
- Query datasources — run queries against Prometheus, Loki, Tempo, Pyroscope, ClickHouse, CloudWatch, and Elasticsearch.
- Read alerting and incidents — inspect alert rules, incidents, OnCall schedules, Sift analyses, and annotations.
- Write, if you allow it — the Cloud server lets you pick read-only or read-write at authorization time; read-write needs an Admin role.
The everyday wins: “why did latency spike at 3pm,” “show me the error-rate panel for checkout,” “what’s firing right now and who’s on call.” All without leaving Claude.
How to set it up
There are two official paths depending on where your Grafana lives:
- Grafana Cloud — add the hosted server at
https://mcp.grafana.com/mcp. It uses an OAuth 2.1 browser flow (no tokens on disk) over streamable HTTP, and it’s supported in Claude Desktop and Claude Code. You’ll need the “Assistant Cloud MCP User” role (Editor+ by default) and to accept the Grafana Assistant terms. It’s still public preview, so expect possible breaking changes. - Self-hosted Grafana — run
grafana/mcp-grafanalocally via uvx, Docker (mcp/grafana), a binary, or Helm, and add it to Claude Desktop or Claude Code. Auth is a Grafana service account token (needs Grafana 9.0+); add--disable-writefor read-only mode.
One thing to get right: Grafana’s legacy API keys are gone — they were fully deprecated and auto-migrated to service account tokens by January 2025. Any guide telling you to “create an API key” is outdated; you create a service account token.
The limits that actually matter
The server is genuinely good at pulling Grafana into a conversation. But its shape is “an analyst you query,” not “an agent that watches.” Three limits define it:
- No triggers, no monitoring. The server only works inside a conversation you start. There’s no “page me when error rate crosses 2%” or “summarize the incident the moment an alert fires.” Grafana’s own alerting fires webhooks, but the MCP server itself doesn’t act on events — you have to be there, prompting.
- Conversation-only. Claude investigates when you ask; it doesn’t sit on your dashboards correlating a new spike on its own. Close the chat and nothing continues.
- Laptop-bound for anything scheduled. The closest thing to “running on its own” is Claude Cowork’s scheduled tasks, which fire on a fixed clock, not on inbox events. That’s not an always-on, event-driven observability agent.
So Claude is great for “help me dig into this spike right now” — and not built for “catch the next one and tell the team before I even look.”
If you want Grafana work that runs on its own: Carly
The moment you want something to happen around Grafana without you in the chat — turn a firing alert into a triaged incident note, email the on-call summary, post a status update when a threshold trips — you’ve crossed past what Claude’s MCP server is for.
That’s where Carly fits. Carly is an AI executive assistant built to act on triggers, not just answer in a chat:
- Fires on events, 24/7, in the cloud — hang a workflow off a Grafana alert webhook (contact points can POST to a URL); your laptop doesn’t need to be awake.
- Connects Grafana to the rest of your stack — take an alert and turn it into an email, a task, a CRM note, or a calendar block in the same workflow.
- Actually sends and updates — drafts and sends email (Gmail and Outlook) with attachments, files and labels, manages tasks, updates your CRM, and records meetings.
- Builds the workflow for you — tell it “when a Grafana alert fires, summarize it and email the on-call engineer” in plain English; it interviews you, then builds it with you. No prompt engineering.
AI agents start at $35/month, and steps in a workflow that don’t use AI run free and unlimited. Carly connects to 200+ tools across 40+ categories — see integrations. Carly natively integrates with Grafana.
Claude vs Carly
| Claude (Grafana MCP) | Carly | |
|---|---|---|
| Query dashboards & datasources | Yes | Yes |
| Read alerts & incidents | Yes | Yes |
| Acts on triggers / events | No | Yes |
| Reacts to a firing alert on its own | No | Yes |
| Works while laptop is closed | No | Yes (cloud) |
| Sends the summary as email | No (Gmail draft-only) | Yes (Gmail + Outlook) |
| Pricing | Pro $20 / Max $100–$200 | AI agents from $35/mo |
Claude’s MCP server is a strong Grafana query tool inside a chat. Carly is a teammate that acts the moment something fires.
Frequently Asked Questions
Does Claude work with Grafana?
Yes, through Grafana’s official MCP server — but it’s a vendor MCP, not an entry in Anthropic’s connectors directory (claude.com/connectors/grafana is a 404). You connect either the hosted Grafana Cloud MCP server (OAuth, public preview) or the self-hosted grafana/mcp-grafana. Once wired up, Claude can query dashboards and datasources and read alerts inside a chat.
How do I connect Claude to Grafana?
For Grafana Cloud, add https://mcp.grafana.com/mcp and authorize with the OAuth browser flow (needs the Assistant Cloud MCP User role). For self-hosted Grafana, run grafana/mcp-grafana locally with a service account token and add it via Claude Desktop or Claude Code. Note that legacy API keys are deprecated — use a service account token.
Can Claude respond to Grafana alerts automatically?
No. The MCP server works inside a conversation you start — it doesn’t act when an alert fires. Grafana can send alert webhooks, but you need an agent platform like Carly to catch them and act 24/7 in the cloud.
Is the Cloud MCP server production-ready?
Not fully — the hosted mcp.grafana.com/mcp server is in public preview as of mid-2026, so breaking changes are possible, and it’s Grafana Cloud only. Self-hosted Grafana users run mcp-grafana locally with a service account token instead.
More: Claude connectors · Claude + Sentry · Can Claude send emails · Claude vs Carly · Claude Cowork alternatives · Best AI agents for productivity
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
